papers/zoo-federated-wildlife/zoo-federated-wildlife.tex

\documentclass[11pt,a4paper]{article}
\usepackage{shared/zoocover}

% ============================================================
% Preamble
% ============================================================
\usepackage[utf8]{inputenc}
\usepackage[T1]{fontenc}
\usepackage{lmodern}
% geometry loaded by zoocover
\usepackage{amsmath,amssymb,amsfonts}
\usepackage{graphicx}
\usepackage{booktabs}
\usepackage{hyperref}
\usepackage{cleveref}
\usepackage{algorithm}
\usepackage{algorithmic}
\usepackage{enumitem}
\usepackage{xcolor}
\usepackage{tikz}
\usetikzlibrary{shapes,arrows,positioning,fit}
\usepackage{float}
\usepackage{subcaption}
\usepackage{multirow}
\usepackage{tabularx}
\usepackage{fancyhdr}
\usepackage{authblk}
\usepackage{abstract}
\usepackage{natbib}
\bibliographystyle{plainnat}

\hypersetup{
    colorlinks=true,
    linkcolor=blue!70!black,
    citecolor=green!50!black,
    urlcolor=blue!80!black
}

\pagestyle{fancy}
\fancyhf{}
\rhead{Antje Worring, Zoo Labs Foundation Research}
\lhead{Federated Learning for Wildlife Monitoring}
\rfoot{Page \thepage}

% ============================================================
\title{\textbf{Federated Learning for Wildlife Monitoring: Privacy-Preserving Multi-Site Analysis} \\[0.5em]
\large Cross-Institutional Data Collaboration with Differential Privacy and Heterogeneous Sensor Fusion}

\author[1]{Antje Worring, Zoo Labs Foundation Research}
\affil[1]{Antje Worring, Zoo Labs Foundation Research, Research Division \\ \texttt{[email protected]}}

\date{January 2025}

\begin{document}

\zoocoverpage

% ============================================================
% Abstract
% ============================================================
\begin{abstract}
\noindent
Wildlife monitoring generates vast quantities of sensitive data across geographically distributed sites managed by independent institutions with heterogeneous sensor deployments and conflicting data-sharing policies.
Training unified machine learning models on this fragmented landscape traditionally requires centralizing raw data---a process that is often legally prohibited, ethically problematic, and logistically impractical.
We present ZooFed, a federated learning framework purpose-built for cross-institutional wildlife monitoring that enables collaborative model training without sharing raw sensor data.
ZooFed introduces three innovations: a \emph{Heterogeneous Sensor Adapter} (HSA) that normalizes inputs from camera traps, acoustic recorders, GPS collars, eDNA samplers, and satellite imagery into a shared feature space compatible with federated aggregation; a \emph{Conservation-Calibrated Differential Privacy} (CCDP) mechanism that allocates privacy budgets based on species sensitivity rather than uniform noise injection, providing stronger protection for endangered species locations; and a \emph{Non-IID Robust Aggregation} (NIRA) algorithm that handles the extreme statistical heterogeneity of ecological data across sites with different species assemblages, climatic conditions, and sensor configurations.
We evaluate ZooFed on a consortium of 18 wildlife monitoring sites across 9 countries, encompassing 4.2 million sensor events from 847 species.
ZooFed achieves 89.4\% species detection accuracy, within 2.1 percentage points of a centralized oracle with full data access, while providing $(\epsilon, \delta)$-differential privacy guarantees with $\epsilon = 2.0$ for common species and $\epsilon = 0.5$ for critically endangered species.
Compared to local-only models, ZooFed improves detection accuracy by 18.7\% on average and by 31.2\% for rare species that benefit most from cross-site knowledge transfer.
\end{abstract}

\vspace{0.5em}
\noindent\textbf{Keywords:} federated learning, wildlife monitoring, differential privacy, sensor fusion, cross-institutional collaboration, conservation AI

% ============================================================
% 1. Introduction
% ============================================================

\input{sections/01-introduction}
\input{sections/02-related-work}
\input{sections/03-problem-formulation}
\input{sections/04-heterogeneous-sensor-adapter}
\input{sections/05-conservation-calibrated-differential-pri}
\input{sections/06-non-iid-robust-aggregation}
\input{sections/07-implementation}
\input{sections/08-evaluation}
\input{sections/09-case-studies}
\input{sections/10-discussion}
\input{sections/99-conclusion}

\end{document}